Travis CI tests on PHP 7.3

[BenMorel]

Travis CI nightly is now PHP 7.4-dev. PHP 7.3 is now available on Travis.

This PR adds Travis tests on PHP 7.3.

[BenMorel]

Never mind, it looks like xdebug is not available yet for PHP 7.3.

A beta version has been released last september, but it's still not readily available on Ubuntu AFAICS.

Let's put this PR on hold until a stable version comes to Ubuntu.

[Majkl578]

Duplicate of unfinished #3323.

[BenMorel]

@Majkl578 Ah, sorry for the dup. Closing this one.

[Majkl578]

If you want, take my commit and you can finish what @morozov suggested, I won't have time for that for next 5 days or so.

[BenMorel]

Actually your commit is based on master, while mine is based on develop (no PHP 7.1 support).

Otherwise they're pretty much equivalent. @morozov are you OK if I apply your suggestions on the commit here?

[morozov]

@BenMorel as long as these are non-breaking changes, we can accept them in master. Sure, you can implement the suggestions from #3323 (comment) here and retarget the pull request.

[BenMorel]

@morozov The problem is that master targets PHP 7.1+, while develop targets 7.2+.

Should I create 2 PRs, one targeting master and one targeting develop?

[morozov]

@BenMorel what is the problem? Regardless of the minimal supported version, we'll support PHP 7.3 in all actively supported branches. No need for two separate pull requests. develop is periodically rebased on top of the latest master.

[BenMorel]

Ah, true, rebasing develop on master should do the job, if there are no conflicts.

OK, I'll pick up @Majkl578's commit and implement your suggestion.

[BenMorel]

OK, so here's what I've done:

• Retargeted the PR on master and picked @Majkl578's commit
• Default versions of sqlite, mysql and mysqli now run only on PHP 7.3 (they were previously running only on PHP 7.2, explicitly excluding 7.1, so I guess it makes sense to increment the version here)
• Kept PHP 7.3 tests only on the latest version of each platform:
  • DB=mysql MYSQL_VERSION=5.7
  • DB=mysqli MYSQL_VERSION=5.7
  • DB=mariadb MARIADB_VERSION=10.3
  • DB=mariadb.mysqli MARIADB_VERSION=10.3
  • DB=pgsql POSTGRESQL_VERSION=11.0
• Removed sqlsrv, pdo_sqlsrv and ibm_db2 tests on PHP 7.3: the sqlsrv and ibm_db2 PECL extensions do not yet support PHP 7.3 and do not compile

@morozov @Majkl578 What do you think?

[morozov]

Looks good so far. A couple of questions:

1. While adding new builds, could you also reduce the number of nightly builds to not increase (or maybe even reduce) the number of builds in total? The pattern I want to achieve is:
   1. For the latest stable PHP version (which is currently 7.2), test all drivers with all platform versions.
   2. All other supported PHP versions (7.1, 7.3, nightly), test each driver with the latest version of the platform.
2. Instead of removing sqlsrv and pdo_sqlsrv, could you try testing their v5.4.0-preview if it's not too difficult to build? It should work.

If the algorithm above makes sense, maybe we could have a script which takes all versions as input and generates the matrix? Just for the sake of discussion and as a reference point.

[BenMorel]

1. Sounds like a good strategy. I'm not sure whether I feel like writing a script to generate the matrix, but I can surely give it a try.
   Note that PHP 7.3 should be GA around December 6, so maybe we can be proactive and consider this one stable right now?
2. You're right, 5.4.0preview works for both sqlsrv and pdo_sqlsrv! Fixed in b85ff57 👍

[morozov]

I'm not sure whether I feel like writing a script to generate the matrix, but I can surely give it a try.

You don't have to. I can give it a try myself. It's just easier to reason about an algorithm and play with the switches when it's implemented as code. If you can just update the matrix following the algorithm, it's fine too.

Note that PHP 7.3 should be GA around December 6, so maybe we can be proactive and consider this one stable right now?

We can do that but I was hoping we'd have all extensions in place by GA (at least sqlsrv and pdo_sqlsrv). Otherwise, we won't have full coverage where we're supposed to have it.

[morozov]

@BenMorel thank you for the thorough explanation. It's most likely not a bug but internal changes in how PHP code is compiled to bytecodes. The change in the coverage is a side effect of that (e.g. like the Xdebug issue #1479).

[morozov]

Please squash the commits and let's get it merged.

[BenMorel]

Can't you squash the commits with 1 button press?

[Majkl578]

We don't squash-merge using GitHub UI because GitHub doesn't create a merge commit (which is what we want) - please squash the commits manually. 😊

[Majkl578]

Also since you did quite some work, you may keep your work as separate commit, or add yourself into Co-Authored-By.

[BenMorel]

I didn't know who the commit would belong to, so I added both of our names (looks like it picked you as the owner, I guess because yours was the first commit?), anyway that seems to work 👍

For my personal knowledge, could you please explain very briefly what's the difference between a merge commit and the squash-and-merge functionality of GitHub?

[morozov]

If everyone is comfortable with the way the work is attributed, I'll merge it in a couple of hours:

Author: Michael Moravec <[email protected]>
Date:   Sun Aug 26 22:04:27 2018 +0200

    CI: Test against PHP 7.3 on Travis
    
    Co-authored-by: Michael Moravec <[email protected]>
    Co-authored-by: Benjamin Morel <[email protected]>

If it's a single commit, based on the amount of work done, I think it'd be fair if it's attributed like:

Author: Benjamin Morel <[email protected]>
Date:   Sun Aug 26 22:04:27 2018 +0200

    CI: Test against PHP 7.3 on Travis
    
    Co-authored-by: Michael Moravec <[email protected]>

What do you say?

[BenMorel]

I don't mind either way!

[morozov]

Done. Thank you @BenMorel and @Majkl578.

[BenMorel]

Thanks 👍

@morozov could you please reply to my comment above?

I didn't get what you mean then. Do you mean a PR with the build matrix as generated by the script (without the root php: and env: entries), but without committing the script itself?

[Majkl578]

For my personal knowledge, could you please explain very briefly what's the difference between a merge commit and the squash-and-merge functionality of GitHub?

The difference is:
a) merge commit exists so it's clear where the commit comes from,
b) multiple commits are allowed,
c) GPG signatures are retained.